Subscribe

Hub/AI Security/Advisories/tool-poisoning-2025

Tool Poisoning Attacks: hidden instructions in MCP tool descriptions

Invariant Labs disclosed that hidden instructions in MCP tool descriptions can make clients exfiltrate files. Mitigations differ by client.

PublishedApr 1, 2025
SeverityNot Rated
CVENone assigned
Statusresearch
AffectedNot listed
Fixed inNot listed
ComponentMCP clients

Sources

Checked 2026-11-04. We copy severity and fix versions from the source and never estimate them.

Affected tools

MCP protocol, SDKs and registryCursorClaude appsSecurity posture matrix

Weekly, free

What changed in AI integrations

One short email when MCP clients, plugin ecosystems, IDE integrations, AI plan prices or security advisories change. Every item is sourced.

You'll finish signing up on beehiiv.